In one line
Copilot App is a purpose-built desktop app for agent-driven development.
Direct multiple agents in parallel, with native ties to GitHub issues, PRs, and CI. You move from writing to directing.
🎯 Generally available (GA) since 2026-06-17. Supports macOS / Windows / Linux. 🆓 Available on all paid Copilot plans (Pro / Pro+ / Business / Enterprise).
What it can do
Built on Copilot CLI, so repositories, branches, issues, and PRs work with no extra setup.
- Parallel workspaces: Run multiple agent sessions at once, each with a dedicated git worktree and branch.
- GitHub integration: Browse issues, start sessions, create / review / merge PRs, and check CI results—all inside the app.
- Model selection: Pick the LLM per session and adjust reasoning effort. BYOK (Bring Your Own Key) lets you use your own provider’s models.
- Canvases: Custom interfaces where people and agents collaborate on the same surface (plan / PR / terminal / browser).
- Automations: Save recurring agent tasks and run them on a schedule or on demand.
- Quick chats: Brainstorm in a conversation mode without creating a branch or workspace.
Session modes
Dial in how much autonomy you hand the agent, in three steps.
| Mode | Behaviour | When to use |
|---|---|---|
| 🤝 Interactive | Collaborate and confirm step by step | Fuzzy specs / you want tight steering |
| 📝 Plan | Agent plans, you approve | Lock down scope and approach first |
| 🚀 Autopilot | Fully autonomous through to code | The task is already well defined |
💡 For early exploration, use a Quick chat to scope before opening a full session—it cuts rework and AI-credit spend.
Parallel workspaces & sandboxes
Each session runs in an isolated git worktree, so files and context never conflict across tasks.
- 🌿 Worktree isolation: A dedicated branch per session. Move to the next task without waiting for one to finish.
- ☁️ Cloud sandboxes (public preview): Run in GitHub-hosted environments—no need to keep your machine awake.
- 🖥️ Local sandboxes: Isolated execution on your own machine.
📝 See Cloud and local sandboxes ↗ for details.
Automations & customization
Repeatable work can be saved and re-run, and you can inject your org’s context into agents.
- 🤖 Automations: Run routine agent tasks on a schedule or on demand.
- 🔧 Customization: Configure global instructions, MCP servers, and agent skills.
- 🕰️ Session history: Use
/chronicleto pull insights from past sessions./chronicle cost tipssurfaces expensive patterns.
/chronicle cost tips # find wasteful patterns across past sessions🔑 BYOK, MCP, and skills share the same extension mechanisms as MCP ↗ and Agent Skills ↗.
★ Killer use case
Directing many agents at once is where Copilot App shines.
Before you leave, spin up five issues, each in its own workspace. Hand the well-defined ones to Autopilot, approve the judgment calls in Plan mode, then review and merge PRs in the order their CI turns green.
Humans decide what to build / which to land; machines handle writing / fixing / verifying. Splitting the boundary lets review-ready PRs flow like a pipeline.
🎯 The IDE is a tool for focusing on one task. Copilot App is the command center for handling many tasks in parallel.
Where it fits among the tools
Same Copilot agent, different role depending on where you direct it from.
| Tool | Home | Strength |
|---|---|---|
| 🖥️ Copilot App | Desktop app | Parallel multi-agent direction · native GitHub integration |
| ⌨️ Copilot CLI | Terminal | Single agent, easy to embed in scripts / CI |
| ☁️ Cloud Agent | GitHub.com | Async execution—just assign an issue |
| 🧩 Partner Agent | Various SDKs / harnesses | Delegate to other harnesses like Claude / Codex |
💡 Copilot App is built on Copilot CLI. Use Copilot CLI ↗ for terminal automation and Cloud Agent ↗ for async runs on GitHub.